Submitted by Benhao Huang 3 Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning Locus Lab 27 2